David F. Percy

Overview

David F. Percy is a statistician at the University of Liverpool (Department of Statistics and Computational Mathematics). His 1992 JRSS-B paper derives Bayesian predictive densities for the Seemingly Unrelated Regressions (SUR) model, establishing that the exact predictive density is analytically intractable in the general case and proposing two approximations: a three-block Gibbs sampler cycling over $(y_{n+1}, \Phi, \beta)$ and a first-order approximation based on the modal Bayes estimate of the precision matrix, which yields a closed-form multivariate normal predictive density.

Key Contributions

• SUR predictive density (Percy 1992): One of the earliest systematic Bayesian treatments of prediction for the SUR model. Identifies the Drèze (1977) intractability barrier and proposes Gibbs sampling and a first-order modal approximation as solutions. The first-order approximation gives a closed-form $N_p[\mu, \Sigma]$ predictive density (eq. 11 of the paper).
• Missing data extension: Extends both approximations to handle missing response components by adding a fourth Gibbs conditional (multivariate normal by partitioning) or by redefining the modal estimate on the observed training data only.

Related

• Seemingly Unrelated Regression
• Gibbs Sampler
• Percy (1992)
• Arnold Zellner